In-network collective operations

Abstract

Examples described herein relate to a switch comprising circuitry configured to for packet communications associated with a collective operation to train machine learning (ML) models: utilize a reliable transport protocol for communications from at least one worker node of the collective operation to a switch, wherein the utilize a reliable transport protocol for communications from at least one worker node of the collective operation to the switch comprises store packet receipt state for per-packet communications from the at least one worker node of the collective operation to the switch and utilize a non-reliable transport protocol by the switch to a device that is to perform aggregation of results, wherein the reliable transport protocol comprises a different protocol than that of the non-reliable transport protocol.

What is claimed is: 
     
         1 . An apparatus comprising:
 an interface and   circuitry coupled to the interface, the circuitry configured to:
 for packet communications associated with a collective operation to train machine learning (ML) models:
 utilize a reliable transport protocol for communications from at least one worker node of the collective operation to a switch, wherein the utilize a reliable transport protocol for communications from at least one worker node of the collective operation to the switch comprises store packet receipt state for per-packet communications from the at least one worker node of the collective operation to the switch and 
 utilize a non-reliable transport protocol by the switch to a device that is to perform aggregation of results, wherein the reliable transport protocol comprises a different protocol than that of the non-reliable transport protocol. 
 
   
     
     
         2 . The apparatus of  claim 1 , wherein the reliable transport protocol comprises one or more of: remote direct memory access (RDMA), InfiniBand, Transmission Control Protocol (TCP), quick UDP Internet Connections (QUIC), or RDMA over Converged Ethernet (RoCE). 
     
     
         3 . The apparatus of  claim 1 , wherein the utilize the reliable transport protocol for communications from at least one worker node of the collective operation to the switch comprises cause transmission of packet receipt acknowledgement (ACK) messages to the at least one worker node based on non-receipt of a packet. 
     
     
         4 . The apparatus of  claim 1 , wherein the non-reliable transport protocol comprises User Datagram Protocol (UDP). 
     
     
         5 . The apparatus of  claim 1 , wherein the utilize a non-reliable transport protocol comprises track receipt of packet transmissions by the switch to a device that is to perform aggregation of results is based on receipt of aggregation of results from a root switch that includes identifiers of contents of the packet transmissions by the switch to the device. 
     
     
         6 . The apparatus of  claim 1 , wherein the circuitry is to cause re-transmission of a packet, associated with an identifier not received from the device, by the switch to the device. 
     
     
         7 . The apparatus of  claim 1 , comprising a switch system on chip (SoC), wherein the switch SoC includes the interface and the circuitry. 
     
     
         8 . At least one non-transitory computer-readable medium comprising instructions stored thereon, that if executed by one or more processors, cause the one or more processors to:
 configure a first network interface device to utilize a reliable transport protocol for communications from at least one worker node of the collective operation to the first network interface device, wherein the utilize a reliable transport protocol for communications from at least one worker node of the collective operation to the first network interface device comprises store packet receipt state for per-packet communications from at least one worker node of the collective operation to the first network interface device and   configure the first network interface device to utilize a non-reliable transport protocol for communications to a second network interface device that is to perform aggregation of results, wherein the reliable transport protocol comprises a different protocol than that of the non-reliable transport protocol.   
     
     
         9 . The non-transitory computer-readable medium of  claim 8 , wherein the reliable transport protocol comprises one or more of: remote direct memory access (RDMA), InfiniBand, Transmission Control Protocol (TCP), quick UDP Internet Connections (QUIC), or RDMA over Converged Ethernet (RoCE). 
     
     
         10 . The non-transitory computer-readable medium of  claim 8 , wherein the utilize a reliable transport protocol for communications from at least one worker node of the collective operation to the first network interface device comprises cause transmission of packet receipt acknowledgement (ACK) messages to the at least one worker node based on non-receipt of a packet. 
     
     
         11 . The non-transitory computer-readable medium of  claim 8 , wherein the non-reliable transport protocol comprises User Datagram Protocol (UDP). 
     
     
         12 . The non-transitory computer-readable medium of  claim 8 , wherein the utilize a non-reliable transport protocol comprises track receipt of packet transmissions by the first network interface device to the second network interface device is based on receipt of aggregation of results from a root switch that includes identifiers of contents of the packet transmissions by the switch to the device. 
     
     
         13 . The non-transitory computer-readable medium of  claim 8 , wherein the first network interface device is to re-transmit a packet, associated with an identifier not received from the second network interface device, to the second network interface device. 
     
     
         14 . The non-transitory computer-readable medium of  claim 8 , comprising instructions stored thereon, that if executed by one or more processors, cause the one or more processors to:
 configure multiple network interface devices to perform collective operations to train machine learning (ML) models by associating logical nodes with the multiple network interface devices.   
     
     
         15 . A method comprising:
 a first network interface device utilizing a reliable transport protocol for communications from at least one worker node of the collective operation to the first network interface device, wherein the utilize reliable transport protocol for communications from at least one worker node of the collective operation to the first network interface device comprises store packet receipt state for per-packet communications from the at least one worker node of the collective operation to the first network interface device and   the first network interface device utilizing a non-reliable transport protocol for communications to a second network interface device that is to perform aggregation of results at least one worker node, wherein the reliable transport protocol comprises a different protocol than that of the non-reliable transport protocol.   
     
     
         16 . The method of  claim 15 , wherein the reliable transport protocol comprises one or more of: remote direct memory access (RDMA), InfiniBand, Transmission Control Protocol (TCP), quick UDP Internet Connections (QUIC), or RDMA over Converged Ethernet (RoCE). 
     
     
         17 . The method of  claim 15 , wherein the utilizing a reliable transport protocol for communications from at least one worker node of the collective operation to the first network interface device comprises transmitting packet receipt acknowledgement (ACK) messages to the at least one worker node based on non-receipt of a packet. 
     
     
         18 . The method of  claim 15 , wherein the non-reliable transport protocol comprises User Datagram Protocol (UDP). 
     
     
         19 . The method of  claim 15 , wherein the utilizing a non-reliable transport protocol comprises tracking receipt of packet transmissions by the first network interface device to the second network interface device based on receipt of aggregation of results from a root switch that includes identifiers of contents of the packet transmissions by the switch to the device. 
     
     
         20 . The method of  claim 15 , comprising:
 the first network interface device causing re-transmission of a packet, associated with an identifier not received from the second network interface device, to the second network interface device.